CPU usage monitoring in AIX servers

Question

In order to monitor the CPU usage in an AIX server I'm using the following script that is executed every 10 mins.

lparstat 2 10 > cpu
usage=$(tail -10 cpu | awk 'BEGIN {sum=0;} {sum+=$4} END{print int(100-sum/10)}')


if [[ $usage -ge 90 ]]; then
# mail the error and cpu file to admin
# displaying this for testing purposes
echo "CPU usage off the charts!!!"
cat cpu
fi

However, in case the CPU usage is above 90% I need to list the top 5 Processes that are using the CPU.

How do I achieve this?

Solution

You can list the top 5 in a fairly readable way by limiting the columns, sorting them with the highest CPU usage first, and then truncating to the first 5 (using head -6, since we also want to include the headers):

ps -eo pcpu,pid,args | sort -k 1 -r | head -6

The output looks something like this:

%CPU   PID COMMAND
 2.0 30531 -bash
 0.0 30673 head -6
 0.0 30672 sort -k 1 -r
 0.0 30671 ps -eo pcpu,pid,args
 0.0 30670 [flush-253:0]

You might also want to look into GNU top's batch mode (-b).

Comments

[debal]

Unfortunately in the AIX system that I'm uing top is unavailable and has topas. ps -eo pcpu,pid,args | sort -rk1 | head -6 does give the output, but the result doesn't match with the result of topas.

[clerksx]

@debal Why must the result match topas?

[debal]

both these commands tend to do the same thing, wouldn't it be surprising if the two results don't match? topas is giving a realtime view of the CPU usage, ps -eo pcpu,pid,args | sort -rk1 | head -6 is giving for a particular point of time, so at that instance it only makes sense if both these commands provide the same result. Which is not the case.

[covener]

That's not exactly the contrast point -- ps gives you a weird process lifetime vs. CPU seconds which is not necessarily indicative of what's causing high CPU "recently". topas, like top, gives you the % over a much short reasonable that is intuitively usually what people want.